Key facts
| Legal fit | Scoped key per matter or system; ethical walls enforced in your gateway |
| API auth | Bearer API keys, scoped per environment and revocable on rotation |
| Workforce identity | SSO with SCIM provisioning for console and admin access |
| Authorization | RBAC with least-privilege scopes per key |
| Retention | Configurable prompt retention; review the DPA for your terms |
| Audit | Authentication, key lifecycle and admin events exportable to SIEM |
| Residency | Region-locked planes plus VPC, on-prem and air-gapped |
| Free plan | plugsky-micro and plugsky-lite, no card; 14-day full-access trial |
TL;DR
- Separate lawyer identity from workload identity; never let a service use a person's account.
- Enforce ethical walls with per-matter keys, namespaces and gateway policy.
- Keep client consent and delegation in the firm's own authorization layer.
- Set prompt retention per workload and document it in the DPA review.
- Export key and inference events to the SIEM so access can be reconstructed matter by matter.
How it works, step by step
- Map each AI workflow to a matter or practice group and decide the isolation boundary before any key is issued.
- Federate the firm's identity provider to Plugsky SSO and automate joiner, mover and leaver flows with SCIM.
- Issue one scoped key per matter, system and environment, held in the firm's secrets manager, and never reuse a key across matters.
- Keep client consent, engagement terms and delegated access in the firm's authorization layer; a broker maps sessions to the correct scoped key.
- Route model calls through the firm's gateway so wall checks, model allow-lists, redaction and logging run before the API.
- Set retention per workload to the minimum the engagement and records policy allow, and confirm those terms in the DPA.
- Stream key, admin and inference events to the SIEM and rehearse wall segregation in an audit exercise.
Try it yourself
Open the API key security checklist →
Two identities and one wall
Legal work has two identity problems. Lawyers and staff need federated sign-in with fast lifecycle changes; that maps to SSO with SCIM and RBAC. Systems — a contract analyser, a discovery summariser — need their own credentials; that maps to scoped keys stored in the firm's secrets manager.
The wall is the third concern, and it is not something the model provider can enforce for you. Ethical segregation lives in your gateway, key allocation and vector namespaces: a matter's prompts, documents and embeddings must never share a namespace or key with another matter. Plugsky supplies the primitives — separate workspaces, scoped keys, region choice — while the firm owns the policy that assigns them.
A practical isolation model
The workable pattern is one workspace or key set per matter, with a naming convention that ties keys to engagement codes, and a broker that refuses to issue a session when the wall check fails. Keep retrieval namespaced per matter, and log the matter identifier alongside model, tokens and latency on every request.
- Least privilege: a summarisation service cannot access a drafting service's key.
- Revocation: closing a matter revokes its keys and freezes its namespaces.
- Retention: set per workload; shorter for exploratory work, longer only where required.
- Audit: key lifecycle and inference events joinable to the matter and the reviewer.
Privilege, retention and honest limits
Privilege and confidentiality are legal concepts, not product features. Plugsky provides scoped authentication, configurable retention, residency choices and audit events; the firm decides what may be processed, by whom, and for how long. Review the terms and DPA with counsel, and document the reasoning so a supervisory review can follow it.
Be precise with clients about endpoint status too: chat completions with streaming, JSON mode and function calling, plus embeddings, are live, while audio, images, files, batch, moderation, fine-tuning and assistants endpoints are labelled coming soon. For plan details and enterprise scoping, see the live pricing page.
Honest comparison
| Concern | Plugsky | Consumer AI assistant | Firm-built platform |
|---|---|---|---|
| Staff identity | SSO with SCIM and RBAC | Personal accounts | Firm IdP integration |
| Matter isolation | Workspaces, scoped keys, namespaced retrieval | Single shared context | Built to firm specification |
| Delegation | Client consent stays in the firm's layer | Not supported | Firm-owned |
| Retention | Configurable per workload, DPA review | Vendor-defined | Firm-defined |
| Audit | Key and inference events to SIEM | Opaque history | Firm logging stack |
Frequently asked questions
How are ethical walls enforced?
In your gateway, key allocation and vector namespaces. Each matter gets its own scoped keys and retrieval namespace, and the broker refuses sessions that cross walls.
Can lawyers share one API key for convenience?
No. A shared key destroys attributability and makes revocation matter-crossing. Issue keys per matter, system and environment, and revoke when the engagement closes.
What about client consent?
Consent, engagement terms and delegated access stay in the firm's authorization layer and client-matter systems. Plugsky receives only the text your policy allows you to send.
How long are prompts retained?
Retention is configurable per workload; set it to the minimum your engagement and records policy allow, and confirm the terms in the DPA.
What audit evidence can a firm produce?
Authentication, key lifecycle, administrative and inference events can be exported to the SIEM, so access can be reconstructed per matter and per reviewer.
Which model tiers make sense for legal work?
Route classification and extraction to plugsky-micro or plugsky-lite, and reserve larger models for reasoning over long documents, all behind one OpenAI-compatible endpoint.
Can we try this on a single matter?
Yes. Start on the free plan with plugsky-micro and plugsky-lite using non-privileged sample data, then move to a paid plan or the 14-day full-access trial when the workflow is proven.